Sample build report · internal lab demo

A mall floor and an office,
from photos alone.

Every delivery comes with a short report like this one: what we measured from your data, what we had to assume, and how the physics checks went — including what didn’t pass. This one is from our own demo, built from 154 existing phone photos. Nobody from our team visited the site.

Full build after several revisions. A free 48 h sample is smaller: one space up to 300 m².

What came out

Inputs and outputs

Photos in
154
Separate assets
642
Asset kits
107
Articulated assets · joints
131 · 431
Lights
533
Triangles at full detail
1.0 M
  • Assets by area: corridor 166 · shops across the atrium 315 · office furniture 43 · office shell 118
  • Every asset has collision shapes and three levels of detail
  • No photos on panels. Walls, shopfronts, furniture and ceiling fixtures are real geometry. Photo pixels appear only on real printed things: sign lettering, posters, door stickers, framed pictures
  • Joints are hinges and slides on fridge doors, pantry drawers, the oven, wall cabinets, storage cabinets, the water dispenser, the copier’s paper trays, shop doors and shutters
  • Lights: downlights, cove lighting, lit signs and shop lighting
  • Reused where possible: common items like chairs came from our existing asset library instead of being modelled again

From the photos

What we measured

Where the same point shows up in two or more photos, we triangulate it. That gives real distances without a scanner.

  • Ceiling heights

    From the light fixtures. The office ceiling has two levels: 2.28 m and 2.43 m.

  • Signs

    Positions and sizes. One shop sign was located from 10 photos, with error at or under 0.6 px.

  • Office layout

    Desks, chairs, pantry, fridge, water dispenser, copier and cabinet heights. The copier and the wall outlet behind it were measured from 8 photos — which moved the copier 1.2 m forward from our first placement, right up to the grey wall with the outlet.

  • Corridor and shopfronts

    Positions of the walkway and shop faces.

  • Colours

    We rendered the twin from each photo’s camera pose and matched floor, desk, tile and wall colours region by region, then applied the camera’s white balance.

Not in the photos

What we assumed — and what we changed

We keep these lists separate so you always know which parts are evidence and which are our best guess.

ASSUMED

  • How deep the shops go beyond what the photos show
  • A storage cabinet seen only in a mirror
  • The upper atrium and parts of the floor above
  • Inside the shutters, drawers and doors

CHANGED ON REQUEST

  • The fridge was closed in every photo. We filled it with 17 scanned food items and added an inside light that turns on when the door opens past 2°
  • Removed a desk phone that sat where the robot puts the bottle down

KNOWN DIFFERENCES

  • Phone HDR brightens some ceilings in the photos, so those look darker in the twin
  • People and some small props were left out

Physics checks

Does it behave, not just look right?

01 · AT REST

Everything free, 3 s of simulation

All joints released and every body simulated for 3 seconds.

Asset pairs overlapping > 2 mm
0
Bodies moving on their own
0
Joints drifting
0
Food items, all with mass
17

02 · EVERY JOINT

Open for ~1 s, then close

Each of the 263 non-shutter joints was driven open and closed on its own. 255 reached at least 85% of their target. The 8 that didn’t are listed below with the reason.

Reached ≥ 85% of target
255 / 263
Water bottles in the fridge door, max slide
11 mm

Bottles in the door shelves slid at most 11 mm while the door swung; none fell. Fruit rides along when the crisper drawer is pulled out.

03 · ROBOT TASK

Open the fridge, take a bottle, set it on a desk

A dual-arm mobile manipulator drives in through the office door, pulls the fridge open by its handle, takes a bottle from the door shelf by friction, pushes the door shut and stands the bottle on a desk. No scripted object motion — only contact.

Task checks passed
24 / 24
Fridge door, max opening
82.1°
Bottle tilt on the desk
0.0025°
Fridge door after closing
0.049°
Contact with the office door
0
Largest tilt of any other item
2.5°

The door-shelf bottles rode the swinging door and stayed within 2 mm of it. The 2.5° is a milk carton that settled 1.6° onto its scanned, slightly uneven base at the start.

The 8 joints that opened less than 85%

PartOpenedWhy
Clinic entrance door79%Heavy door, motor torque capped at 60 N·m, so 1 s wasn’t enough. Given 3 s it reaches 98% — and then touches a banner stand that is also in the photos.
Marble-front entrance door44%Heavy door, same 60 N·m cap. Reaches its target given 3 s.
Fridge crisper drawer9%Blocked by the closed fridge door — as it would be in real life.
Storage cabinet A, door 222%Hits a hanging sign we measured from the photos.
Storage cabinet B, door 116%Hits the corridor wall.
Storage cabinet B, door 216%Hits the corridor wall.
Copier bypass tray69%Hits the edge of the mirror behind it. The photos show the tray open, so our estimated hinge position is slightly off.
Task chair swivel84%Hits the mirror behind it and a computer on the desk.

Same camera, same pose

Photo vs twin

Left: the real photo. Right: the twin, rendered from that photo’s exact camera pose and field of view.

Real photo of the mall corridorThe twin from the same pose
Corridor
Real photo of the officeThe twin from the same pose
Office
Real photo of a shop entranceThe twin from the same pose
Shop door
Real photo of shops across the atriumThe twin from the same pose
Across the atrium

Want this report for your space?

Send photos, a floor plan, a SLAM map or CAD. You get a sim-ready sample in 48 h, with a report like this one. First space free.